Sno Station Skills gives Claude Code and Codex 31 team habits

A coding agent that finishes a task at 3am and says the tests are green now has a second one that reads the result in a fresh session from the other vendor and reports what would break in real use. Sno Station Skills is a set of 31 skills, released on 1 October 2026 under Apache-2.0, that install into Claude Code and Codex and give both agents the working habits of a team. The GitHub repository had 401 stars by 16 October.

The package installs with one command per agent: sno setup --harness claude for Claude Code, sno setup --harness codex for Codex CLI. OpenClaw and Hermes Agent are also supported. The skills are grouped into six categories: peer review and audit, build and rollout, communication and handoff, reporting and visibility, delegation and work orders, retrospective and improvement.

What the skills change at the console

Three skills carry most of the pitch. peer-review runs the reviewer in a fresh session with none of the author's conversation and prefers the other vendor's command-line tool. With only one tool installed it still runs, in a fresh session, and the first line of the report says it was a same-vendor review. less-is-more climbs a ladder before writing code: is this needed at all, is it already in the repository, does the standard library do it, does the platform do it, does an installed dependency do it. It cuts one-implementation interfaces, config knobs for values that never change, and scaffolding "for later". owner-intent-audit checks whether a plan still matches what was asked for: a condition that became unconditional, an example that became the whole boundary, a prohibition that quietly disappeared.

The overnight handoff

rotate-agent and subscription-quota-check work together. Arm rotate-agent before bed. When the working agent's subscription drops to the threshold, it writes a handover brief while it still can, wakes a fresh agent from the other vendor, and hands the task over with the exact commit to continue from. The brief is verified before a single file is touched. away-brief writes a one-page summary of the night, 60 lines or fewer, when the user comes back: what got done, what is stuck and why, what waits for an answer, and what it cost in quota.

The nightly retrospective

rem-reflect reads the day's Claude Code and Codex sessions, finds the mistakes that repeat, and proposes refinements to the skills in use. Every change needs the user to approve it, which is the whole point: a skill pack that edits itself at 3am is not what the project does. The install also carries charter, which asks for a written list of success checks against which "done" is answered pass or fail by a real run, and agentic-walkthrough, which walks a program hop by hop to catch a job that reports success and touched nothing.

The practical shape is that one Claude Code or Codex subscription is still one subscription, and the package does not change that. What changes is that a long run started at midnight does not stop dead at 3am, a change does not land on the author's own say-so, and a mistake made on Monday has a filed note by Tuesday. For a two-person team paying for both Claude Code and Codex, that is the first open-source version of the cross-vendor review the two agents could already do by hand, with the hand step removed.

Source

sno-ai/sno-station-skills on GitHub, Apache-2.0 licensed. The parent project is Sno Station.